Imagine robots that can play soccer (football) at the level of the World Cup championships. For researchers in artificial intelligence, such an event would be tantamount to—and possibly even surpass—that moment in 1997 when IBM’s Deep Blue supercomputer defeated then-world champion Garry Kasparov in chess.
The challenges are daunting. Autonomous, athletically capable humanoids that act together as a unit would require not just highly advanced software (the intellectual component) but also highly advanced hardware (the physical component). By sharing knowledge and codes, and developing and testing technologies together, AI designers hope to realize this vision.
Launched in 1993, the RoboCup international robot soccer competition (also known as the Robot World Cup Initiative) provides a platform for AI and robotics researchers to test their developments, work together, spur each other on, and create research breakthroughs. It is a competition in the best sense of the word—the kind that facilitates cooperation.
In his essay “Robot Soccer,” University of New South Wales computer science and engineering professor Claude Sammut describes the different levels of play, pointing out that the robotic soccer fields are smaller (and virtual in some low-level competitions), and the rules much simpler than in soccer played by humans. Currently, there are only three robots per team, as compared to eleven in human play. Sammut writes: “As the robots and their programming have become more sophisticated, the rules of the game, including field size and number of players, have been made tougher to encourage progress.”
French company Aldebaran Robotics’ humanoid Nao is the model of robot currently in use in the RoboCup. While still relatively basic, these humanoid robots use color cameras as their primary sensors (not unlike HAL in 2001: A Space Odyssey), operate autonomously (as opposed to being remote-controlled), and can communicate with each other wirelessly.
Sammut stresses that soccer is only a means to an end—not an end in itself. “In addition to soccer playing, the competition also includes leagues for urban search and rescue and for robotic helpers at home,” he writes. He emphasizes that soccer is good for developing the fundamentals that will be necessary for these and many other tasks. The basics include “perceiving” their surroundings, interpreting constantly changing situations, making quick decisions based on those situations, and then acting on them, adjusting tactics as necessary. The AI units must also be able to transmit information back and forth.
Whereas soccer fields always conform to the same basic grid layout and boast the same landmarks (goal posts, for example), less-structured environments present greater challenges. For example, a house or apartment and the possessions it contains (which can act as landmarks) may not change much over time, but it is more complex to move about in. It is harder still for an AI program to map a completely unfamiliar urban environment without any immediately identifiable landmarks. In search and rescue situations, “the robot has to simultaneously map its environment while reacting to and interacting with the surroundings,” Sammut writes. And off the soccer field, AI units must interact with actual people—not just other AI units.
Despite the challenges, little by little, progress is being made each year. And if the participants and organizers meet their stated goals, then expect a team of fully autonomous humanoid robots to show no mercy against their opponents in the actual World Cup in 2050.—Aaron M. Cohen
Sources: Claude Sammut, “Robot Soccer,” Wiley Interdisciplinary Reviews: Cognitive Science, Volume 1, Issue 6, http://wires.wiley.com/WileyCDA.
RoboCup, www.robocup.org.